The Ramp AI Index for June 2026 puts Anthropic at 41% of US business AI spending share. OpenAI is at 32.3%. Ramp is a corporate expense-management platform tracking actual card transactions — not a survey of stated preferences or projected usage — across more than 50,000 US companies.

A year ago, only 9% of businesses paid for Anthropic. That's 32 percentage points in twelve months, during which OpenAI's business spending share grew by 0.3 points. Not 0.3x. 0.3 points. Roughly flat.

The crossover itself happened in May. The Axios report dated May 13 caught it first; the June data has the gap widening further.

Source spread

• Ramp AI Index — May/June 2026 — builder. The primary source: actual corporate card spend across 50,000+ US companies. This is spending share, not headcount or installation count — a methodological choice that has meaningful implications.
• VentureBeat — Anthropic finally beat OpenAI — builder. Growth trajectory analysis plus three risks to the lead: government action (now live), IDC data discrepancy, OpenAI's revenue moat.
• Axios — Anthropic overtakes OpenAI in workplace AI adoption — builder. First major account of the crossover from the May data.
• SaaStr — velocity curves tell a different story — skeptic. OpenAI's counter-case: higher absolute penetration, deeper existing relationships, and consumer brand advantages that corporate card data doesn't capture.

Pros & cons

Why the Ramp data matters:

• Actual spend, not stated intent. "64% of IT decision-makers plan to increase AI investment" is noise. "$41 of every $100 businesses spend on AI goes to Anthropic tools" is a transaction. The signal quality is categorically different.
• The 32-point gain in twelve months is the number. Not the current share — the velocity. Going from 9% to 41% in one year, while the incumbent barely moves, is the kind of shift that redraws categories if it continues.
• What's actually driving it? Enterprise buyers naming reasons consistently: instruction-following reliability on complex prompts, long-context handling without degradation, and lower hallucination rates on structured workflows. Those are sticky advantages in production environments where errors have real costs.

Why the Ramp data is also incomplete:

• Spending share is not revenue share. OpenAI almost certainly still generates more total revenue, and reportedly is on pace for higher 2026 annual revenue than Anthropic. Winning 41% of the per-company spend at large US businesses is real. Winning on total economic output — not established by this data.
• The IDC enterprise survey from roughly the same period tells a different story: only 19% of organizations report extensive use of Claude across the enterprise, compared to higher figures for OpenAI and Google tools. Ramp and IDC are measuring different things — spending concentration at companies that pay versus broad organizational penetration. Both numbers are true; neither is the whole story.
• Correlation with the Fable 5 situation. As of June 12, Anthropic's two newest and most capable models are offline globally due to a US Commerce Department export control directive. That doesn't erase the Ramp data, but it makes the July Ramp index reading more important than usual.

Samwise's take

What builders need to know

• Ramp data = spending share, not revenue share. Anthropic leads in "percentage of US companies that pay for Anthropic." OpenAI likely still leads in total revenue. These are different metrics. Know which one you're citing.
• The velocity is the actual story. 9% to 41% in twelve months is a market-movement signal, not a product-feature signal. Enterprise buyers are switching, not just adding. That changes the competitive landscape for products built on top of these APIs.
• The Fable 5 outage is now an enterprise procurement risk factor. Every enterprise AI RFP for the next six months will have a version of this question: "What happens to our access if the government issues an export control directive?" Have an answer ready.
• Watch the July Ramp index. The June data was collected before June 12. The July data will be the first post-Fable-5-shutdown reading. If Anthropic's spending share holds or grows, the outage was a speed bump. If it drops, something structurally changed.
• Build multi-provider routing if you haven't. Not because Anthropic's reliability is in question — their track record on uptime is strong. But because the June 12 event proved that provider access can be interrupted from outside the vendor-customer relationship. Treat it the same way you'd treat a single-region architecture: fine until it isn't.

Further reading

• Ramp AI Index — May/June 2026 — the primary data source, methodology explained
• VentureBeat — Anthropic finally beat OpenAI in business AI adoption — three risks to the lead, including the now-active government risk
• Axios — Anthropic overtakes OpenAI in workplace AI adoption — first major account of the crossover
• SaaStr — velocity curve analysis — the OpenAI counter-case
• Anthropic — Statement on the US government directive to suspend Fable 5 and Mythos 5 — the enterprise risk context